[WIP] Polish previous test and added two more

The tests results are quite cute:
```
case	line	dash	bash	mksh	ash	osh
  0	  4	FAIL	pass	FAIL	pass	FAIL	signals aren't reset in subshells if not trapping
  1	 16	pass	pass	pass	pass	pass	signals are reset in subshells if trapping
  2	 41	pass	pass	pass	pass	FAIL	ignored signals aren't reset in subshells
```
More spec tests on the behavior of signal ignoring can be done.
Both `dash` and `mksh` aren't POSIX compliant due to them failing `case 0`.
In the case of `osh`, `case 2` is failed because it doesn't give special
treatment to `trap '' USR1`. All the other sells set the signals to
SIG_IGN, which is different from an empty handler. This also affects
the behavior of `exec`, but that's painful to test.

Author: rolfan-ish

spec/builtin-trap.test.sh

@@ -1,17 +1,62 @@
 ## compare_shells: dash bash mksh ash
 ## oils_failures_allowed: 3
 
-#### resest signal inside subshell
+#### signals aren't reset in subshells if not trapping
+trap foo USR1
+
+# just 'trap' doesn't modify the current traps
+if test -n "$(trap)"; then
+  echo ok
+fi
+
+## STDOUT:
+ok
+## END
+
+#### signals are reset in subshells if trapping
+trap foo USR1
+subshell_result=$(
+  trap foo USR2
+  trap
+)
+trap - USR1
+trap foo USR2
+
+# We need to store in order to avoid using subshell expansion.
+# We would be comparing subshell against subshell instead of
+# parent shell against subshell otherwise.
+tmpfile=$(mktemp)
+trap > "$tmpfile"
+shell_result=$(cat "$tmpfile")
+rm "$tmpfile"
+
+if test "$shell_result" = "$subshell_result"; then
+  echo ok
+fi
+
+## STDOUT:
+ok
+## END
+
+#### ignored signals aren't reset in subshells
 trap '' USR1
-trap foo INT
-(
+subshell_result=$(
   trap '' USR2
+  trap - USR2
   trap
 )
 
+tmpfile=$(mktemp)
+trap > "$tmpfile"
+shell_result=$(cat "$tmpfile")
+rm "$tmpfile"
+
+if test "$shell_result" = "$subshell_result"; then
+  echo ok
+fi
+
 ## STDOUT:
-trap -- '' SIGUSR1
-trap -- '' SIGUSR2 
+ok
 ## END
 
 #### trap accepts/ignores --